Song searching method and device, storage medium and electronic equipment

Abstract

The invention relates to the field of data recall, in particular to a song searching method, a song searching device, a storage medium and electronic equipment. The song searching method comprises the following steps: extracting hot key words based on user generated contents of songs, and creating an association relationship between the hot key words and the songs; and in response to a search term input by a user, determining a target song query result according to the search term and the association relationship between the hotspot keyword and the song. According to the song searching method provided by the invention, the content generated by the user can be mined to form association between the song and the external hotspot, and the song searching experience of the user is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of data recall, and particularly to a song search method, a song search device, a storage medium, and an electronic device. Background Art

[0002] With the rapid development of the Internet, especially the popularization of mobile Internet, people can quickly and conveniently obtain various hot information, and major Apps will recommend hot content of interest to users.

[0003] Currently, for music-related hotspots, the main source is from news platforms. The data is irregular, there is a lot of noise data, the timeliness is lagging, and a large amount of operation manpower is required to analyze and label the associated music content.

[0004] It should be noted that the information disclosed in the above background art section is only used to enhance the understanding of the background of the present disclosure, and thus may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0005] The present disclosure provides a song search method, a song search device, a storage medium, and an electronic device, aiming to mine user-generated content to form an association between songs and external hotspots, and improve the user's song search experience.

[0006] Other features and advantages of the present disclosure will become apparent through the following detailed description, or be learned in part through the practice of the present disclosure.

[0007] According to one aspect of the present disclosure, there is provided a song search method, including:

[0008] Extracting hot keywords based on user-generated content of songs, and creating an association relationship between the hot keywords and the songs;

[0009] Responding to a search term input by a user, and determining a target song query result according to the search term and the association relationship between the hot keywords and the songs.

[0010] Optionally, when the association relationship is a song set corresponding to a hot keyword, the creating of the association relationship between the hot keyword and the song includes:

[0011] Creating a first mapping relationship between the song and the hot keyword;

[0012] And performing an inverted index according to the first mapping relationship to obtain a second mapping relationship between the hot keyword and the song;

[0013] Obtaining the song set corresponding to the hot keyword according to the second mapping relationship.

[0014] Optionally, when the association relationship is a list of hot keywords corresponding to a song, creating the association relationship between the hot keywords and the song includes:

[0015] For a song, aggregating the hot keywords corresponding to the song according to the time dimension to obtain at least one keyword aggregation result;

[0016] Calculating the association degree value between each hot keyword and the song based on each keyword aggregation result, and counting the frequency value of each hot keyword based on the keyword aggregation result;

[0017] Creating a list of hot keywords corresponding to the song according to the association degree value and the frequency value of each hot keyword.

[0018] Optionally, extracting hot keywords based on the user-generated content of the song includes:

[0019] Performing data cleaning on the user-generated content to construct content data;

[0020] Determining candidate keywords from the content data and scoring the candidate keywords to obtain candidate keyword scores;

[0021] Selecting a preset number of the candidate keywords as the hot keywords based on the candidate keyword scores.

[0022] Optionally, scoring the candidate keywords to obtain candidate keyword scores includes:

[0023] Extracting the feature values of the candidate keywords and calculating the weight values of the candidate keywords based on the feature values; the feature values include smoothed word frequency feature values, word context correlation feature values, and word difference feature values;

[0024] Combining the candidate keywords into combined candidate words, and calculating the weight values of the combined candidate words according to the weight values of each candidate keyword, the inverse document frequency feature values and the word frequency feature values of the combined candidate words;

[0025] Scoring according to the weight values of the combined candidate words to obtain the candidate keyword scores.

[0026] Optionally, determining the target song query result according to the search term and the association relationship between the hot keywords and the song includes:

[0027] Querying the target hot keywords that match the search term, and querying the target song set that matches the target hot keywords according to the song set corresponding to the hot keywords in the association relationship;

[0028] When the target song set includes one song, use the target song as the target song query result;

[0029] When the target song set includes multiple songs, based on the list of hot keywords corresponding to the songs in the association relationship, perform a fusion score on each target song to obtain a song sorting result, which is used as the target song query result.

[0030] Optionally, the performing a fusion score on each target song based on the list of hot keywords corresponding to the songs in the association relationship to obtain a song sorting result, which is used as the target song query result, includes:

[0031] Obtain a first sequence value of the hot keyword based on the correlation value of the hot keyword in the list of hot keywords; and

[0032] Obtain a second sequence value of the hot keyword based on the frequency value of the hot keyword in the list of hot keywords;

[0033] Score the target song according to the first sequence value and the second sequence value to obtain the score of the target song;

[0034] Traverse all the target songs, sort the scores of each target song to obtain the song sorting result.

[0035] According to a second aspect of the present disclosure, there is provided a song search device, including:

[0036] An association module, configured to extract hot keywords based on user-generated content of songs and create an association relationship between the hot keywords and the songs;

[0037] A query module, configured to respond to a search term input by a user and determine a target song query result according to the search term and the association relationship between the hot keywords and the songs.

[0038] According to a third aspect of the present disclosure, there is provided a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the song search method in the above embodiments.

[0039] According to a fourth aspect of the embodiments of the present disclosure, there is provided an electronic device, characterized in that it includes: one or more processors; a storage device for storing one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the song search method in the above embodiments.

[0040] The exemplary embodiments of the present disclosure may have the following partial or all beneficial effects:

[0041] In the technical solutions provided by some embodiments of the present disclosure, hot keywords are extracted based on user-generated content related to songs, and an association relationship is created between the hot keywords and the songs. Furthermore, a song recall service is provided by using the pre-constructed hot keywords and the association relationship. On the one hand, by mining hot keywords based on user-generated content, music hotspots associated with songs can be discovered at low cost, and at the same time, the timeliness of the hot keywords can be improved. On the other hand, the data used for data recall includes the association relationship between the pre-mined hot keywords and the songs, enabling the relationship between the hot keywords and the songs to be considered simultaneously during data recall, thereby improving the accuracy of song query results and further enhancing the user's search experience.

[0042] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and do not limit the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure. Obviously, the drawings in the following description are only some embodiments of the present disclosure, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts. In the drawings:

[0044] Figure 1 Schematically showing a flowchart of a song search method in an exemplary embodiment of the present disclosure;

[0045] Figure 2 Schematically showing a logical diagram of a method for extracting hot keywords in an exemplary embodiment of the present disclosure;

[0046] Figure 3 Schematically showing a schematic diagram of a filtering rule in an exemplary embodiment of the present disclosure;

[0047] Figure 4 Schematically showing a schematic diagram of a filtering model in an exemplary embodiment of the present disclosure;

[0048] Figure 5 Schematically showing a schematic diagram of content normalization processing in an exemplary embodiment of the present disclosure;

[0049] Figure 6 Schematically showing a logical diagram of extracting hot keywords based on content data in an exemplary embodiment of the present disclosure;

[0050] Figure 7 Schematically showing a schematic diagram of determining candidate keywords in an exemplary embodiment of the present disclosure;

[0051] Figure 8 A schematic diagram showing the extraction results of hot keywords in an exemplary embodiment of the present disclosure;

[0052] Figure 9 A schematic diagram showing hot keywords under different aggregation methods in an exemplary embodiment of the present disclosure;

[0053] Figure 10 A schematic logic flowchart showing the construction of a song recall service in an exemplary embodiment of the present disclosure;

[0054] Figure 11 A schematic flowchart showing data recall based on hot keywords in an exemplary embodiment of the present disclosure;

[0055] Figure 12 A schematic diagram of a graphical user interface for song recall in an exemplary embodiment of the present disclosure;

[0056] Figure 13 A schematic diagram showing the composition of a song search device in an exemplary embodiment of the present disclosure;

[0057] Figure 14 A schematic diagram showing the structure of a computer system of an electronic device in an exemplary embodiment of the present disclosure. Detailed implementation manners

[0058] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be more complete and comprehensive, and will fully convey the concept of the example embodiments to those skilled in the art.

[0059] In addition, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of the embodiments of the present disclosure. However, those skilled in the art will realize that the technical solutions of the present disclosure can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, well-known methods, devices, implementations, or operations are not shown or described in detail to avoid obscuring aspects of the present disclosure.

[0060] The block diagrams shown in the drawings are only functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0061] The flowcharts shown in the accompanying drawings are only illustrative and not necessarily include all the content and operations / steps, nor are they necessarily executed in the order described. For example, some operations / steps can be decomposed, while some operations / steps can be combined or partially combined. Therefore, the actual execution order may change according to the actual situation.

[0062] With the rapid development of the Internet, especially the popularization of mobile Internet, people can quickly and conveniently obtain various hot information, and major apps will also recommend hot content that users are interested in to users. However, how to find the desired music content based on hot words has always been a difficult problem.

[0063] Currently, for the task of hot content mining, the following three common technologies are available:

[0064] The main solution of Technology One is to select authoritative and timely Weibo media accounts from Sina Weibo and use web crawler technology to obtain the corresponding Weibo texts. Then, based on the method of constructing an opinion knowledge graph for Weibo hot events, for each hot event crawled, the corresponding people and institutions, as well as the relationships among the three, a network diagram is constructed to form an opinion knowledge graph, which provides an important reference basis for maintaining the corporate image and formulating crisis public relations strategies.

[0065] However, this situation has the following disadvantages: 1. Simply analyzing the release results of some Weibo accounts may have subjective biases in the event content, and the hot range covered by a single platform is also very limited. 2. Only analyze and identify people, institutions, and the relationships between them. It is impossible to analyze the music and entertainment entities associated with hot events, such as songs, albums, etc. 3. Only a relationship graph is formed, and it is impossible to conduct a detailed descriptive analysis of the identified hot events. 4. It is impossible to analyze the impact of hot content in a specific field, such as the impact of hot events on music, artists, etc., that is, whether a hot event can make a song or an artist popular.

[0066] Technology Two discloses a hot news prediction method and system based on Weibo data analysis. Collect news reports from mainstream news websites and the Weibo user response information they trigger on Weibo; establish a hot news prediction model for Weibo texts to judge whether subsequent news will become hot. The goal is to comprehensively analyze the trend of news reports in Weibo topics and predict whether the news will become an opinion hot spot, so as to solve the problem of early prediction of hot news.

[0067] This situation has the following disadvantages: 1. Only relying on the topic of Weibo text for trend prediction may not have a high confidence level. 2. There is no discussion on how to filter and process the noise data on Weibo, such as spam and advertisements, which may affect the accuracy of the prediction. 3. Only the hot trends are predicted, and the associated entities of the hot topics are not analyzed. Moreover, for the same hot topic, it may have different impacts in different fields.

[0068] Technique three provides a search-based hot topic discovery method, aiming to improve the accuracy and effectiveness of the hot topic list. The method includes obtaining the hot topic keywords and their generation time, the maximum search volume in the current period and the previous N periods, calculating the heat value, and sorting and displaying the hot topics accordingly. Factors such as the current search volume, historical peak search volume, and time decay are considered, and the current state and trend of the hot topic are reflected through the heat index. In addition, it also involves using a click-through rate model to update the hot topic list to fit the user click situation.

[0069] Although this situation improves the timeliness of the hot topic list by combining the trend information of the search, it also has the following problems: 1. There is no analysis of the entities and keyword analysis for the hot topics, and the specific entities of the hot topics cannot be associated. 2. There is no classification of the hot topics, so it cannot be applied to the music platform.

[0070] Based on the above introduction, for the current music-related hot topics, the main source method is from news platforms, where the data is irregular, there is a lot of noise data, the timeliness is lagging, and a large amount of operational manpower is required for analysis, annotation, and association with music content.

[0071] Therefore, aiming at the disadvantages in the existing technology, in order to discover the music hot topics associated with songs at low cost and better recommend the hot topic search results to users, the present disclosure provides a song search method. Based on the real-time music comments under the song and the included playlist data, it mines the hot topic keywords associated with the song, scores and sorts these hot topic keywords. At the same time, it aggregates the hot topic keywords with multiple time granularities and constructs a hot topic keyword index according to the scoring importance. For the user's search of relevant hot topic words, it gives the associated songs and provides an explanatory text for the association, so as to improve the retrieval experience of hot topic-related content.

[0072] Before introducing the technical solution of the present disclosure, some terms are explained.

[0073] UGC: UGC is the abbreviation of "User Generated Content", which means user-generated content, referring to the text, pictures, videos, etc. originally created and shared by users on the Internet platform.

[0074] Song Review: A song review is the written feedback of users on a specific song or music work, expressing personal feelings, evaluations, or interpretations. These reviews usually appear on music streaming platforms, social media, or music sales websites, aiming to share first impressions, emotional experiences, or professional analyses of the song. Reviews not only help other users decide whether to listen to a certain song but also provide feedback to artists, promoting interaction and discussion in the music community.

[0075] Playlist: A playlist is a list created by users on a music playback platform that contains a series of songs, used to organize and share personal favorite music. Users can create playlists according to themes such as mood, activity, or music style, or discover new songs based on the recommendations of other users. Playlists facilitate users to manage and play music, and at the same time promote social interaction as users can follow and share playlists with each other.

[0076] The implementation details of the technical solutions of the embodiments of the present disclosure are elaborated in detail below.

[0077] Figure 1 A flowchart of a song search method in an exemplary embodiment of the present disclosure is schematically shown. As Figure 1 shown, the song search method includes steps S101 to S103:

[0078] Step S101, extracting hot keywords based on user-generated content of the song and creating an association relationship between the hot keywords and the song;

[0079] Step S103, in response to a search term input by the user, determining a target song query result according to the search term and the association relationship between the hot keywords and the song.

[0080] In the technical solutions provided by some embodiments of the present disclosure, hot keywords are extracted based on user-generated content of the song, and an association relationship between the hot keywords and the song is created, and then a song recall service is provided by using the pre-constructed hot keywords and association relationships. On the one hand, mining hot keywords based on user-generated content can discover music hotspots associated with the song at low cost and improve the timeliness of hot keywords; on the other hand, the data used for data recall includes the association relationship between the pre-mined hot keywords and the song, enabling the relationship between the hot keywords and the song to be considered simultaneously during data recall, thereby improving the accuracy of the song query result and further enhancing the user's search experience.

[0081] Next, each step of the song search method in this exemplary embodiment will be described in more detail with reference to the accompanying drawings and embodiments.

[0082] In step S101, hot keywords are extracted based on the user-generated content of the song, and the association relationship between the hot keywords and the song is created.

[0083] Specifically, it is first necessary to pre-construct the hot keywords of the song.

[0084] Hot keywords generally refer to the words that can represent the theme or core content extracted from a piece of text. In the music scenario, they refer to the words closely related to the song.

[0085] For a song, the user-generated content corresponding to the song can be extracted first. User-generated content, i.e., UGC, refers to the text, pictures, videos, etc. originally created and shared by users on the Internet platform. In the music field, it can specifically include the comments of users on a certain song, the titles of the playlists summarized, etc.

[0086] After the UGC is extracted, keyword extraction is performed according to the content dimension of the UGC. The content dimension refers to different categories of UGC, such as comment data, playlist data, etc. In this disclosure, in order to improve the keyword mining efficiency, the UGC to be mined can be first represented in the data form of a standard document, and then the standard document is used for extraction.

[0087] In the embodiments of this disclosure, the method for extracting hot keywords specifically includes:

[0088] Step 1: Clean the user-generated content to construct content data;

[0089] Step 2: Determine candidate keywords from the content data and score the candidate keywords to obtain candidate keyword scores;

[0090] Step 3: Select a preset number of the candidate keywords as the hot keywords based on the candidate keyword scores.

[0091] Figure 2 Schematically shows a logical diagram of a method for extracting hot keywords in an exemplary embodiment of this disclosure. As Figure 2 shown, the specific process of extracting hot keywords based on the UGC corresponding to the song is as follows: First, aggregate the UGC content associated with the song, then after filtering by rules and models, normalize the content to construct a document, and finally use the document for keyword extraction, and the extraction results in a specific application scenario are given.

[0092] Next, refer to Figure 2 to describe each of the above steps in detail.

[0093] In step 1, clean the user-generated content to construct content data.

[0094] Among them, first, the UGC construction content data corresponding to the song can be a document corresponding to one song, or multiple documents corresponding to different content dimensions of UGC. For example, a document aggregating song comments, or a document aggregating song playlist titles. The present disclosure does not make specific limitations here. Refer to Figure 2 As shown, documents can be constructed for comments and playlists respectively.

[0095] When constructing the content data, data cleaning is required because the UGC content of users may be meaningless garbage content and may contain prohibited content. It is necessary to clean the data through filtering rules or filtering models. The user-generated content is cleaned according to the preset filtering rules. The filtering rules are simply set but cannot cover all situations; using a pre-trained text classification model to clean the user-generated content has a good filtering effect but is complex to set. In specific use, they can be used separately or in combination. For example, first use the filtering rules and then use the filtering model.

[0096] Figure 3 A schematic diagram showing a filtering rule in an exemplary embodiment of the present disclosure is schematically shown. For cleaning data using filtering rules, rule-based data filtering can be performed by summarizing some common garbage content patterns. The specific filtering rules can be referred to Figure 3 .

[0097] Figure 4 A schematic diagram showing a filtering model in an exemplary embodiment of the present disclosure is schematically shown. For cleaning data using the filtering model, a text classification model can be trained and learned to filter out garbage content.

[0098] The text classification model can be trained using positive samples corresponding to keywords without filtering rules and with available content, hard positive samples including keywords corresponding to filtering rules but with available content, and negative samples including keywords corresponding to filtering rules and with unavailable content.

[0099] Specifically, appropriate samples need to be introduced during model training. For positive samples, we select high-praise and high-quality comments. At the same time, to make the model more generalizable and robust, hard positive samples are introduced, that is, some samples that contain common words in the filtering rules but are high-quality comments. Negative samples are mainly some prohibited comments, as well as some address comments and playlists that hit the rules.

[0100] The chinese-roberta-wwm-ext pre-trained in Chinese can be used on the model and fine-tuned on this basis. After experiments, the accuracy rate of model classification can reach 96.1%. Among them, chinese-roberta-wwm-ext is a Chinese pre-trained model optimized based on the RoBERTa architecture, which was released in 2019. It uses the Whole Word Masking technology to mask the whole word, improving the model's understanding ability of Chinese texts. It is applicable to NLP tasks such as text classification, named entity recognition, and sentiment analysis.

[0101] It should be noted that in addition to using the above-mentioned chinese-roberta-wwm-ext for text classification, some other large language models such as qwen can also be used for filtering, and the present disclosure does not make any limitations here.

[0102] In an embodiment of the present disclosure, after cleaning the data, the content text can also be normalized. Figure 5 Schematically shows a schematic diagram of a content normalization process in an exemplary embodiment of the present disclosure, as Figure 5 shown, the normalization process includes, for example, normalization, cropping, merging, etc.

[0103] After constructing the content data, hot keywords are extracted using the content data. Figure 6 Schematically shows a logical diagram of extracting hot keywords based on content data in an exemplary embodiment of the present disclosure. As Figure 6 shown, first, the document is input, and then in the first step, candidate keywords are determined from the content data; in the second step, the candidate keywords are scored to obtain the candidate keyword scores; in the third step, the Top-K is selected as the hot keywords of the song according to the candidate keyword scores, and the extraction results in specific scenarios are also given.

[0104] It should be noted that extracting keywords from content data composed of UGC is very different from extracting keywords from ordinary documents. The main differences are two points: First, there is no central theme, which means that the method based on feature representation cannot be used. The basic idea of ordinary documents is to first extract the theme vector of the document, and then convert the keywords in the document into vectors to calculate the similarity, which greatly increases the difficulty of keyword extraction; Second, there is no labeled data, so only unsupervised methods can be used for extraction.

[0105] Regarding the first point above, the present disclosure provides a statistical method for keyword extraction, which is optimized based on the yake! algorithm for the music scenario; regarding the second point, search and external word lists are introduced to solve the problems of a large number of Internet terms and new words in UGC content and the great difficulty of keyword extraction.

[0106] Among them, YAKE! is a lightweight unsupervised automatic keyword extraction method that selects the most important keywords in the text based on text statistical features extracted from a single document. This method does not require training on a specific document set and does not rely on dictionaries, external corpora, text size, language, or domain. YAKE! can handle keywords containing internal stop words (such as "Game of Thrones") and has a high precision. It also has linear scalability, can adapt to documents of different lengths, and is not limited by the frequency of occurrence of candidate keywords.

[0107] In step S203, candidate keywords are determined from the content data, and scores are assigned to the candidate keywords to obtain candidate keyword scores.

[0108] Specifically, refer to Figure 6 the first step in

[0109] Figure 7 schematically shows a schematic diagram of determining candidate keywords in an exemplary embodiment of the present disclosure. As Figure 7 shown, for a content data, meaningful phrase candidates to be sorted are segmented from the content data. First, word segmentation is performed. For example, entity information such as songs or artists in a music library can be segmented. Hundreds of billions of texts can be processed within half an hour from single-machine word segmentation to Spark distributed word segmentation, which can solve the problem of large document content. Then, NER (Named Entity Recognition) is recognized. Search and external word lists can be introduced to recognize long sentences, such as poems, lyrics, or some popular internet memes. Then they are combined into N-gram combinations to provide high-quality search terms. In the traditional part-of-speech word combination method, word combination can also be combined with the search scenario. For example, entity word combinations with meanings in the user search scenario can be combined into phrases, and the longest phrase can be composed of 20-gram. Finally, the final candidate keywords are obtained after stop word filtering.

[0110] Among them, Spark is an open-source distributed computing system that provides a fast and general-purpose cluster computing platform. It supports a variety of data processing tasks, including batch processing, real-time stream processing, machine learning, and graph computing. Spark is famous for its speed, can perform calculations in memory, and also supports disk storage. Its rich API supports multiple programming languages, such as Scala, Java, Python, and R. The resilient distributed dataset (RDD) and DataFrame abstractions of Spark make large-scale data processing simple and efficient.

[0111] N-gram is a concept in natural language processing and computational linguistics, used to predict the probability of consecutive items in text. It is based on the assumption that the probability of an item's occurrence is related to the previous several items. The N-gram model predicts the next item by statistically counting and evaluating the number of times given words, terms, or n-grams appear consecutively. For example, in a bigram model where N = 2, the model considers the probability of two consecutive words. The N-gram model is widely applied in fields such as language models, text analysis, and speech recognition.

[0112] Reference Figure 6 In the second step of , score the candidate keywords to obtain the candidate keyword scores.

[0113] The advantage of the yake! algorithm compared to the traditional tfidf is mainly that it can naturally support scoring for extracting combined words, and at the same time, the feature calculation is more comprehensive. However, the yake! algorithm also has problems. For example, it is not suitable for Chinese, and it is a keyword extraction algorithm for traditional documents, so it needs to be adapted to the Chinese music scenario. Therefore, the existing yake! algorithm needs to be improved, and the specific improvement content is as follows: First, make Chinese adaptation for yake!. Second, remove the features of case and word position for UGC content adaptation. Third, the combined word scoring formula is optimized for the Chinese scenario, and the IDF score within the scenario is introduced.

[0114] In an embodiment of the present disclosure, the specific process of scoring the candidate keywords using the improved yake! algorithm to obtain the candidate keyword scores is as follows:

[0115] First, extract the feature values of the candidate keywords, and calculate the weight values of the candidate keywords based on the feature values; the feature values include smoothed word frequency feature values, word context relevance feature values, and word difference feature values.

[0116] The smoothed word frequency feature can solve the bias problem of high-frequency words, and the specific formula is as follows:

[0117]

[0118] The word context relevance feature, that is, the more combinations with other words, the lower the score, and the specific formula is as follows:

[0119]

[0121] The word difference feature indicates how many different sentences a word appears in, and the specific formula is as follows:

[0122]

[0123] The calculation of the above features is the same as that of the prior art, so no more explanations will be given here for the publicly disclosed content.

[0124] Then, combine the candidate keywords into combined candidate words, and calculate the weight value of the combined candidate words according to the weight values of the candidate keywords, the inverse document frequency feature values and the word frequency feature values of the combined candidate words. Score individual words, and the specific formula is as follows:

[0125]

[0126] Finally, score according to the weight value of the combined candidate words to obtain the score of the candidate keywords. Score the combined words based on individual words, and introduce the IDF score in the scenario. The specific formula is as follows:

[0127]

[0128] Among them, IDF is the abbreviation of Inverse Document Frequency, that is, inverse document frequency, and IDF measures the general importance of words.

[0129] On the Chinese CSL benchmark (benchmark test or benchmark dataset) which is the Chinese dataset 10k, compare the P (Precision), R (Recall), and F1 (harmonic mean of precision and recall) values of the TF-IDF algorithm, the textrank algorithm, the yake! algorithm, and the improved yake! algorithm provided by the present disclosure. Finally, the P value of the TF-IDF algorithm is 0.1174, the R value is 0.1237, and the F1 value is 0.1135; the P value of the textrank algorithm is 0.1049, the R value is 0.1084, and the F1 value is 0.1002; the P value of the yake! algorithm is 0.1116, the R value is 0.0867, and the F1 value is 0.0887; the P value of the improved yake! algorithm is 0.1396, the R value is 0.1577, and the F1 value is 0.1407. As can be seen from the above, the improved yake! algorithm has significantly improved performance on the Chinese dataset 10k.

[0130] Based on the above method, by optimizing the yake! algorithm, adapting to the Chinese music scenario, and supporting the scoring of combined words, the accuracy and efficiency of hot keyword extraction can be improved.

[0131] In step S205, select a preset number of the candidate keywords as the hot keywords based on the scores of the candidate keywords.

[0132] Reference Figure 6In the third step, after determining the candidate keywords, rank the scores for each candidate keyword, reduce the redundancy in the selected key phrases, and then select the top-k, that is, the top k ones.

[0133] Figure 8 A schematic diagram showing the extraction result of a hot keyword in an exemplary embodiment of the present disclosure. Specifically, control the proportion of the co-occurrence times before and after the combined word to the occurrence of a single word, and at the same time judge whether there is repetition, such as "wula wula wula" and "wula", and remove the duplicates. The example song is She-Jami Soul / Mr.Bang / The keywords are displayed in the form of a word cloud, and the final extraction result is as Figure 8 shown in Figure 8.

[0134] Based on the above method, by mining the UGC content associated with music to obtain hot keywords, the association between songs and external hotspots can be formed. Such an association relationship can be precipitated as the assets of the songs, providing a good data basis for subsequent distribution, recall, recommendation, and operation.

[0135] In an embodiment of the present disclosure, when the association relationship is a song set corresponding to a hot keyword, creating the association relationship between the hot keyword and the song includes: creating a first mapping relationship between the song and the hot keyword; and performing an inverted index according to the first mapping relationship to obtain a second mapping relationship between the hot keyword and the song; obtaining the song set corresponding to the hot keyword according to the second mapping relationship.

[0136] Specifically, after obtaining the hot keyword by aggregating the UGC of the song, a mapping of song → hot keyword can be created, and then through the method of inverted index, a mapping result of hot keyword → song list can be formed. Finally, all the mapping relationships are summarized to obtain all the songs corresponding to the same hot keyword, and then a song set corresponding to the hot keyword is formed. By constructing an efficient keyword index through inverted index, fast recall and sorting can be supported.

[0137] Among them, the inverted index is an index data structure used to store the mapping relationship between words in a document and the documents where the words are located. It allows the system to quickly retrieve documents containing specific words. In an inverted index, each unique word corresponds to a list that contains references to all the documents in which the word appears. This index structure is widely used in full-text search engines such as Elasticsearch and Apache Lucene to achieve fast text search functions.

[0138] At the same time, in the process of inverted mapping, the corresponding song ancillary information can be associated, including artist name, album name, etc., especially the original UGC content used to build the association relationship, because it is rich in explanations of the association between songs and hot spots.

[0139] In one embodiment of the present disclosure, when the association relationship is a list of hot keywords corresponding to a song, the creation of an association relationship between the hot keywords and the song includes: for one of the songs, aggregating the hot keywords corresponding to the song according to the time dimension to obtain at least one keyword aggregation result; calculating the association value between each of the hot keywords and the song based on each of the keyword aggregation results, and counting the frequency value of each of the hot keywords based on the keyword aggregation results; and creating a list of hot keywords corresponding to the song according to the association value and the frequency value of each of the hot keywords.

[0140] As time points change, the hot topics associated with songs are also constantly changing, and the more recent the content, the more new hot topics will suddenly appear. For example, the Foshan Dianhan meme is about a host who suddenly became popular on Douyin. Because he looks like Zhang Han, and the BGM played during the live broadcast is the song with id=5307435, there is a correlation between the two. Therefore, in order to distinguish the keyword content at different time points, keywords of different dimensions can be aggregated, and then the aggregated keywords can be analyzed.

[0141] Specifically, first, for a song, the hot keywords corresponding to the song are aggregated according to the time dimension to obtain at least one keyword aggregation result. The time dimension is the time span, such as 30 days, 90 days, half a year or all historical data.

[0142] It should be noted that the extraction is not done only once, but multiple extractions are performed according to different time dimensions to obtain multiple copies of target content data, and then the multiple copies of target content data are analyzed, such as the text comments corresponding to the song within 30 days, and the playlist name corresponding to the song within half a year. In this way, short-term and long-term hot spots associated with the same song can be obtained.

[0143] Figure 9 The schematic diagram schematically shows a hot keyword in a different aggregation mode in an exemplary embodiment of the present disclosure. Figure 9 As shown, for song id = 5307435, the hot keywords corresponding to the comments on the song within 90 days and the hot keywords corresponding to the comments on the song in all time are aggregated as shown in the figure. It can be seen from the above that in the past 90 days, the song is highly correlated with "Foshan Dianhan", and looking at all the data, the song is highly correlated with the term "happy worship".

[0144] Then, based on the keyword aggregation results of each, calculate the association degree values between each of the hot keywords and the song, and count the frequency values of each of the hot keywords based on the keyword aggregation results.

[0145] Specifically, in order to highlight the impact of timeliness on hot keywords, when finally calculating the association degree value of each keyword under a song, it is necessary to fuse the keyword aggregation results of different types. The keyword aggregation results under different aggregation methods will have a greater impact on the association degree value.

[0146] In order to rank the recently popular keywords at the front and downweight the words that rarely appear recently, the scores of the hot keywords in the keyword aggregation result with the longest time dimension can be calculated first. This score calculation can be performed in the same way as the scoring of the candidate keywords described above; then different weights are assigned to other time dimensions. For example, the weight of 30 days is higher than that of 60 days. The number of times the hot keywords appear in the keyword aggregation results under other time dimensions and the weights corresponding to the time dimensions are used to adjust the scores of the hot keywords, and finally the association degree values of the hot keywords under the song are obtained.

[0147] Taking the fusion of the first keyword aggregation result corresponding to the comments in the most recent 90 days and the second keyword aggregation result corresponding to the comments in the entire life cycle of the song as an example. Since only the two aggregation results are fused and analyzed, on the basis of the second keyword aggregation result, only the first keyword aggregation result is used for adjustment, so there is no need to configure weights separately.

[0148] In the first keyword aggregation result, count the number of times a keyword appears, denoted as show 90d As the recent popularity coefficient of the keyword; in the second keyword aggregation result, score the keyword according to the improved yake! algorithm described above to obtain the score score.

[0149] The score of the fused keyword is:

[0150]

[0151] Based on the above method, by aggregating UGC content in different time periods, short-term and long-term hot keywords are obtained, thereby improving the timeliness and comprehensiveness of hot spot recognition.

[0152] At the same time, the frequency values of the hot keywords under the song, that is, the number of occurrences, can also be counted. The frequency of occurrence of the hot keywords can also reflect the degree of association between the keyword and the song to a certain extent.

[0153] Finally, create a list of hot keywords corresponding to the song according to the association degree values and the frequency values of each of the hot keywords.

[0154] Based on the above method, keywords are aggregated at multiple time granularities and multiple content granularities to create correlations between hot keywords and songs, which is conducive to more accurate mining of short-term and long-term hot spots corresponding to the song, and makes the recently popular keywords ranked higher, while the words that have rarely appeared recently are downgraded to a certain extent, making the user's search results more timely and more in line with the current music environment.

[0155] Figure 10 The following schematically shows a logic flow chart of constructing a song recall service in an exemplary embodiment of the present disclosure. Figure 10 As shown, in order to build an engine for song recall service, it is roughly divided into the following three stages.

[0156] Phase 1: mining hot keywords associated with songs, that is, mining hot keywords associated with songs through UGC content such as comments and playlists on the site;

[0157] Phase 2: Multi-granularity hot keyword aggregation of songs, that is, aggregating UGC content such as comments and playlists in different time periods to obtain short-term and long-term hot topics associated with songs;

[0158] Phase 3: Build a song hot keyword engine to provide recall services, aggregate the mined keywords, and build an inverted index so that keywords can be associated with corresponding hot songs and their ancillary information, as well as the source of the corresponding hot association as an explanation.

[0159] After pre-extracting hot keywords and creating the association between the hot keywords and songs, recall service can be provided.

[0160] In step S103, in response to the search word input by the user, the target song query result is determined according to the search word and the association relationship between the hot keywords and the song.

[0161] Figure 11 The following is a schematic diagram of a process for recalling data based on hot keywords in an exemplary embodiment of the present disclosure. Figure 11 As shown in the figure, the correlation between hot keywords and songs is constructed as an engine by analyzing UGC content in advance, and the correlation can be used to recall data based on the constructed engine. Recall means to filter out resources related to the user input from the candidate resource pool according to certain conditions based on the information input by the user.

[0162] In one embodiment of the present disclosure, the specific process of step S103 is as follows: Query the target hot keywords that match the search term, and according to the song set corresponding to the hot keywords in the association relationship, query the target song set that matches the target hot keywords; when the target song set includes one song, use the target song as the target song query result; when the target song set includes multiple songs, perform fusion scoring on each of the target songs according to the list of hot keywords corresponding to the songs in the association relationship to obtain a song sorting result, which is used as the target song query result.

[0163] Specifically, when the user inputs a search term, first determine the target hot keywords corresponding to the search term. Here, the definition of fuzzy search can be adopted, such as automatically replacing misspelled words, synonyms, and complementing the search term, etc., to match the corresponding target hot keywords.

[0164] Then, according to the song list corresponding to the hot keywords in the association relationship, obtain the target songs that match the target songs. If there is only one song in the song list, then the song can be directly returned as the target song query result; however, if there are multiple songs in the song list, at this time, the song list is not directly used as the target song query result, but it is also necessary to re-perform fusion scoring and sorting on each song in the song list.

[0165] In one embodiment of the present disclosure, the performing fusion scoring on each of the target songs according to the association relationship to obtain a song sorting result specifically includes:

[0166] Step 1: Obtain the first sequence value of the hot keyword based on the association degree value of the hot keyword in the list of hot keywords; and

[0167] Step 2: Obtain the second sequence value of the hot keyword based on the frequency value of the hot keyword in the list of hot keywords;

[0168] Step 3: Score the target song according to the first sequence value and the second sequence value to obtain the score of the target song;

[0169] Step 4: Traverse all the target songs, and sort the scores of each target song to obtain the song sorting result.

[0170] The above steps will be described in detail below.

[0171] In step 1, obtain the first sequence value of the hot keyword based on the association degree value of the hot keyword in the list of hot keywords.

[0172] Specifically, according to the association relationship, a list of hot keywords corresponding to each song can be extracted, which records the association degree values corresponding to each hot keyword after the fusion aggregation result. Taking the association degree value as the sorting factor and searching the entire list of hot keywords, the sequence value rank of the association degree value of this hot keyword in the list of hot keywords can be obtained. s2q , as the first sequence value.

[0173] In step two, the second sequence value of the hot keyword is obtained based on the frequency value of the hot keyword in the list of hot keywords.

[0174] Similar to step one, taking the frequency value as the sorting factor and searching the entire list of hot keywords, the sequence value rank of the frequency value of this hot keyword in the list of hot keywords can be obtained. q2s , as the second sequence value.

[0175] In step three, the target song is scored according to the first sequence value and the second sequence value to obtain the score of the target song.

[0176] Specifically, considering both the first sequence value rank s2q and the second sequence value rank q2s for RRF fusion scoring in two dimensions. RRF stands for Reciprocal Rank Fusion, which is an algorithm for fusing multiple result sets. It combines the result sets of different relevance metrics into one result set. RRF works by assigning reciprocal rank scores to the documents in each result set. This method can obtain high-quality search results without adjusting the relationship between different relevance metrics. The specific formula is as follows:

[0177] score =1 / (rank s2q +50 )+1 / (rank q2 +50) (7)

[0178] Then, after sorting the recalled songs according to the result of the fusion scoring, they are displayed to the user on the APP side.

[0179] Based on the above method, using the RRF algorithm to fuse the sorting positions of the association degree value and the occurrence times of keywords in the song for comprehensive scoring and sorting can improve the accuracy and relevance of the recall results.

[0180] It should be noted that the RRF fusion scoring in this disclosure is only an exemplary illustration. In fact, other methods can also be used for scoring according to the first sequence value and the second sequence value, such as a normalized linear weighted scoring scheme.

[0181] Figure 12Schematically shown is a schematic diagram of a graphical user interface for song recall in an exemplary embodiment of the present disclosure. For example, when a user searches for a search term such as [Foshan Dianhan], relevant songs can be accurately recalled and an explanation for the recall can be given, that is, the original GUC content for constructing the association relationship, including the explanation for the association between the song and the hot topic. Refer to Figure 12 as shown in (a). When the user searches for [Mid-Autumn Festival], the displayed results are as Figure 12 shown in (b).

[0182] Through Figure 12 it can be seen that the user's search experience is better and the recalled results more meet the diverse needs of users. Experiments were conducted on search recall. The effective click-through rate of the searched song module increased by +3.37%, the per capita effective duration increased by +2.79%, and the secondary retention rate of the search scenario increased significantly by +0.31%.

[0183] Based on the above method, an efficient and accurate song recall system is provided, which supports users to input hot keywords for searching, improving user experience and satisfaction. And through algorithm optimization and index construction, fast and accurate association of hot keywords and song recall are realized.

[0184] In an embodiment of the present disclosure, the above method for automatically mining song hot topics based on UGC content can also be migrated to other scenarios, such as mining podcast hot topics, community hot topics, etc., and has wide applicability.

[0185] In an embodiment of the present disclosure, the query results of the target songs to be recalled can also be provided to AIGC for playlist compilation. AIGC, i.e., artificial intelligence-generated content, refers to the process of automatically generating or editing content using artificial intelligence technology, including text, images, audio, and video, etc. It learns from a large amount of data and imitates the creation methods of humans to improve the efficiency and quality of content production. AIGC technology is widely used in fields such as news writing, art creation, and game development. The created playlist is played 1,000,000 times per day, and the relevance score of the manual evaluation of the recall is 4.25 / 5 points, and the quality score is 4.95 / 5 points.

[0186] Figure 13 Schematically shown is a schematic diagram of the composition of a song search device in an exemplary embodiment of the present disclosure, as Figure 13 shown. The song search device 1300 may include a creation module 1301 and a query module 1303. Among them:

[0187] The creation module 1301 is used to extract hot keywords based on the user-generated content of the song and create an association relationship between the hot keywords and the song;

[0188] A query module 1303, configured to, in response to a search term input by a user, determine a target song query result according to the search term and the association relationship between the hot keywords and songs.

[0189] According to an exemplary embodiment of the present disclosure, the creation module 1301 includes a first association unit, configured to, when the association relationship is a set of songs corresponding to a hot keyword, create a first mapping relationship between the song and the hot keyword; and perform an inverted index according to the first mapping relationship to obtain a second mapping relationship between the hot keyword and the song; and obtain a set of songs corresponding to the hot keyword according to the second mapping relationship.

[0190] According to an exemplary embodiment of the present disclosure, the creation module 1301 further includes a second association unit, configured to, when the association relationship is a list of hot keywords corresponding to a song, for a song, aggregate the hot keywords corresponding to the song in a time dimension to obtain at least one keyword aggregation result; calculate an association degree value between each hot keyword and the song based on each keyword aggregation result, and calculate a frequency value of each hot keyword based on the keyword aggregation result; create a list of hot keywords corresponding to the song according to the association degree value and the frequency value of each hot keyword.

[0191] According to an exemplary embodiment of the present disclosure, the creation module 1301 further includes an extraction unit, configured to perform data cleaning on the user-generated content to construct content data; determine candidate keywords from the content data, and score the candidate keywords to obtain candidate keyword scores; select a preset number of the candidate keywords as the hot keywords based on the candidate keyword scores.

[0192] According to an exemplary embodiment of the present disclosure, the extraction unit is further configured to extract a feature value of the candidate keyword, and calculate a weight value of the candidate keyword based on the feature value; the feature value includes a smoothed term frequency feature value, a term context relevance feature value, and a term difference feature value; combine the candidate keywords into combined candidate words, and calculate a weight value of the combined candidate word according to the weight value of each candidate keyword, the inverse document frequency feature value and the term frequency feature value of the combined candidate word; score the candidate keyword according to the weight value of the combined candidate word to obtain the candidate keyword score.

[0193] According to an exemplary embodiment of the present disclosure, the query module 1303 is configured to query target hot keywords that match the search term, and query a set of target songs that match the target hot keywords according to the set of songs corresponding to the hot keywords in the association relationship; when the set of target songs includes one song, use the target song as the target song query result; when the set of target songs includes multiple songs, perform a fusion score on each of the target songs according to the list of hot keywords corresponding to the songs in the association relationship to obtain a song sorting result, which is used as the target song query result.

[0194] According to an exemplary embodiment of the present disclosure, the query module 1303 is further configured to obtain a first sequence value of the hot keyword based on the association degree value of the hot keyword in the list of hot keywords; and obtain a second sequence value of the hot keyword based on the frequency value of the hot keyword in the list of hot keywords; score the target song according to the first sequence value and the second sequence value to obtain the score of the target song; traverse all the target songs, and sort the scores of each target song to obtain the song sorting result.

[0195] The specific details of each module in the above song search device 1300 have been described in detail in the corresponding song search method, so they will not be elaborated here.

[0196] It should be noted that although several modules or units of a device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of the two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0197] In an exemplary embodiment of the present disclosure, there is also provided a storage medium capable of implementing the above method. It can use a portable compact disc read-only memory (CD-ROM) and include program code, and can run on a terminal device, such as a mobile phone. However, the program product of the present disclosure is not limited to this. In this document, a readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, device, or device.

[0198] In an exemplary embodiment of the present disclosure, there is also provided an electronic device capable of implementing the above method. Figure 14 Schematically shows a structural diagram of a computer system of an electronic device in an exemplary embodiment of the present disclosure.

[0199] It should be noted that Figure 14The computer system 1400 of the illustrated electronic device is merely an example and should not impose any limitation on the functions and usage scope of the embodiments of the present disclosure.

[0200] As Figure 14 shown, the computer system 1400 includes a central processing unit (CPU) 1401, which can perform various appropriate actions and processes according to the programs stored in the read-only memory (ROM) 1402 or the programs loaded from the storage section 1408 into the random access memory (RAM) 1403. In the RAM 1403, various programs and data required for system operation are also stored. The CPU 1401, ROM 1402, and RAM 1403 are connected to each other via a bus 1404. An input / output (I / O) interface 1405 is also connected to the bus 1404.

[0201] The following components are connected to the I / O interface 1405: an input section 1406 including a keyboard, a mouse, etc.; an output section 1407 including, for example, a cathode ray tube (CRT), a liquid crystal display (LCD), etc. and a speaker, etc.; a storage section 1408 including a hard disk, etc.; and a communication section 1409 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 1409 performs communication processing via a network such as the Internet. A drive 1410 is also connected to the I / O interface 1405 as required. A removable medium 1411, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1410 as required so that the computer program read from it can be installed into the storage section 1408 as required.

[0202] Specifically, according to the embodiments of the present disclosure, the processes described below with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for performing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from the network through the communication section 1409, and / or installed from the removable medium 1411. When the computer program is executed by the central processing unit (CPU) 1401, various functions defined in the system of the present disclosure are executed.

[0203] It should be noted that the computer-readable medium shown in the embodiments of the present disclosure can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of a computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, apparatus, or device. In the present disclosure, a computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, in which computer-readable program code is carried. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, and this computer-readable medium can send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0204] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram can represent a module, a program segment, or a part of code, and the above module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks can occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks shown can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, as well as the combination of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0205] The units involved in the embodiments described in this disclosure can be implemented in software or in hardware, and the described units can also be provided in a processor. Among them, the names of these units do not, in some cases, constitute a limitation on the unit itself.

[0206] As another aspect, the present disclosure also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or may exist alone without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the one or more programs are executed by an electronic device, the electronic device implements the methods described in the above embodiments.

[0207] It should be noted that although several modules or units of the device for action execution are mentioned in the above detailed description, such a division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of the two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0208] Through the description of the above embodiments, those skilled in the art can easily understand that the example embodiments described herein can be implemented by software or by a combination of software and necessary hardware. Therefore, the technical solutions according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (such as a personal computer, a server, a touch terminal, or a network device, etc.) to execute the methods according to the embodiments of the present disclosure.

[0209] After considering the specification and practicing the invention disclosed herein, those skilled in the art will readily conceive of other embodiments of the present disclosure. The present disclosure is intended to cover any variations, uses, or adaptations of the present disclosure, which follow the general principles of the present disclosure and include well-known knowledge or conventional technical means in the technical field not disclosed by the present disclosure.

[0210] It should be understood that the present disclosure is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present disclosure is only limited by the appended claims.

Claims

1. A song search method, characterized in that: include: Extract hot keywords based on the user-generated content of the song and create associations between the hot keywords and the song; In response to a search word input by a user, a target song query result is determined according to the search word and the association relationship between the hot keywords and the song.

2. The song search method according to claim 1, characterized in that: When the association relationship is a set of songs corresponding to the hot keywords, the step of creating an association relationship between the hot keywords and the songs includes: Creating a first mapping relationship between the song and the hot keyword; And performing inverted indexing according to the first mapping relationship to obtain a second mapping relationship between the hot keywords and the songs; The song set corresponding to the hot keyword is obtained according to the second mapping relationship.

3. The song search method according to claim 1, characterized in that: When the association relationship is a list of hot keywords corresponding to songs, the step of creating an association relationship between hot keywords and songs includes: For the song, aggregating the hot keywords corresponding to the song according to the time dimension to obtain at least one keyword aggregation result; Calculating the relevance value of each of the hot keywords and the song based on the keyword aggregation results, and counting the frequency value of each of the hot keywords based on the keyword aggregation results; A hot keyword list corresponding to the song is created according to the association value and the frequency value of each hot keyword.

4. The song search method according to claim 1, characterized in that: The method of extracting hot keywords based on user-generated content of songs includes: Performing data cleaning on the user-generated content to construct content data; Determining candidate keywords from the content data, and scoring the candidate keywords to obtain candidate keyword scores; A preset number of the candidate keywords are selected as the hot keywords based on the candidate keyword scores.

5. The song search method according to claim 4, characterized in that: Scoring the candidate keywords to obtain candidate keyword scores includes: Extracting feature values ​​of the candidate keywords, and calculating weight values ​​of the candidate keywords based on the feature values; the feature values ​​include smoothed word frequency feature values, word context relevance feature values, and word difference feature values; Combining the candidate keywords into combined candidate words, and calculating the weight value of the combined candidate words according to the weight value of each candidate keyword, the inverse document frequency feature value and the word frequency feature value of the combined candidate words; The candidate keyword score is obtained by scoring according to the weight value of the combined candidate words.

6. The song search method according to claim 1, characterized in that: The step of determining the target song query result according to the search term and the association relationship between the hot keywords and the song includes: Querying the target hot keywords matching the search term, and querying the target song set matching the target hot keywords according to the song set corresponding to the hot keywords in the association relationship; When the target song set includes a song, taking the target song as the target song query result; When the target song set includes multiple songs, each of the target songs is scored based on the hot keyword list corresponding to the songs in the association relationship to obtain a song ranking result as the target song query result.

7. The song search method according to claim 6, characterized in that: The step of fusing and scoring the target songs according to the hot keyword list corresponding to the songs in the association relationship to obtain a song ranking result as the target song query result includes: Acquire a first sequence value of the hot keyword based on the association value of the hot keyword in the hot keyword list; and Acquire a second sequence value of the hot keyword based on the frequency value of the hot keyword in the hot keyword list; Scoring the target song according to the first sequence value and the second sequence value to obtain a score of the target song; All the target songs are traversed, and the scores of the target songs are sorted to obtain the song sorting result.

8. A song search device, characterized in that: include: Creating a module for extracting hot keywords based on user-generated content of songs and creating associations between hot keywords and songs; The query module is used to respond to the search words input by the user and determine the target song query results according to the search words and the association relationship between the hot keywords and the songs.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the song search method as described in any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, enables the one or more processors to implement the song search method as described in any one of claims 1 to 7.

Images